An entropy penalized approach for stochastic control problems. Complete version

Thibaut Bourdais, Nadia Oudjane and Francesco Russo
submitted
Publication type:
Paper in peer-reviewed journals
arXiv:
assets/images/icons/icon_arxiv.png 2309.01534
Keywords :
Stochastic control problem; Optimization; Donsker-Varadhan representation; Relative entropy; Exponential twisting.
Abstract:
In this paper, we propose an alternative technique to dynamic programming for solving stochastic control problems. We consider a weak formulation that is written as an optimization (minimization) problem on the space of probabilities. We then propose a regularized version of this problem obtained by splitting the minimization variables and penalizing the entropy between the two probabilities to be optimized. We show that the regularized problem provides a good approximation of the original problem when the weight of the entropy regularization term is large enough. Moreover, the regularized problem has the advantage of giving rise to optimization problems that are easy to solve in each of the two optimization variables when the other is fixed. We take advantage of this property to propose an alternating optimization algorithm whose convergence to the infimum of the regularized problem is shown. The relevance of this approach is illustrated by solving a high-dimensional stochastic control problem aimed at controlling consumption in electrical systems.
BibTeX:
@article{Bou-Oud-Rus-2200-1,
    author={Thibaut Bourdais and Nadia Oudjane and Francesco Russo },
    title={An entropy penalized approach for stochastic control problems. 
           Complete version },
    year={submitted },
    month={9},
}